Paid Subscriber Perks

E-commerce hasn’t had a paradigm shift since its inception. You have a category page, a product page, and checkout. You search, and then you narrow down with filters. There is an incredible amount of friction baked into the ways we shop that people underestimate. The average shopper spends 15+ hours, across 12 different websites, over the course of 79 days trying to find and decide on a product. That’s crazy, right? And what’s even more unbelievable is that even after LLMs and ChatGPT launched, the time it takes for a shopper to make a purchase decision is increasing, not decreasing.

The idea for Onton started in 2020 when Alex (Co-Founder and CTO of Onton) experienced this problem firsthand. He was looking for a gray couch with wood trim on the bottom, in a mid-century style, for under $800. It took him 30 hours to find it, scrolling through various websites and opening hundreds of tabs.

Our mission is to bring that 79-day shopping journey down to less than 1. Helping consumers make decisions they love, instantly.

We could have turned our product into an LLM wrapper and moved much faster, but we found that for all the impressive feats of large language models (LLMs), they end up compounding the pain points of e-commerce. LLMs are generating exabytes of unstructured, unverified data daily, and the tools at shoppers’ disposal are neither prepared nor incentivized to deal with it. The computer science adage of “garbage in, garbage out” applies: you can’t shortcut your way to better search by adding an “intuition” layer on top of bad data.

Collectively, we’ve spent over a year building our own graph database, vector database, and self-learning AI to solve some of the fundamental problems in e-commerce decision-making. Most importantly, so that we can always provide an accurate, trustworthy response.

There are two: Internally, its conversion rate. By giving people great results plus the information they need to make a decision, we’ve seen our conversion rate hit 3x the e-commerce benchmark and almost 10x the existing tools in the market.

Externally, there are a lot of studies that verify the struggles with e-commerce, but a recent one that continues to provide a north star is that more shoppers than ever are abandoning their carts because of analysis paralysis.

LLMs are causing the mass extinction of the data that they rely on to provide great results (both for training and for search). This is felt acutely in e-commerce. Think of all the incredible expert opinions and reviews from Consumer Reports, Wirecutter, personal blogs, individual product reviews, etc. There is no incentive to create this type of content anymore because products like ChatGPT and Perplexity are summarizing it. In addition, the content that is still being created is now a majority AI-Generated.

Without that data, it’s nearly impossible to rank the best quality products. To help consumers make decisions, we need this type of data. Consumer products that incentivize and create a space on a platform for this type of content to be created will have a major advantage (Reddit is a great example).

Sometimes, it takes a maximally viable product to meet demand. Obvious examples of this would be self-driving vehicles (where you need a certain level of safety to drive), but I also think this applies to purely software products, too. Maybe the most relevant example is Figma, where designers needed a certain critical mass of tools to see adoption (it took them 4 years to see substantial growth).

One of the toughest games in building a startup is knowing if your product needs that or not and persevering until that point.